421F39f091003a43dc7c22e364c

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XLII-3/W8, 2019 Gi4DM 2019 – GeoInformation for Disaster Management, 3–6 September 2019, Prague, Czech Republic

FLOOD ANALYSIS WITH REMOTE SENSING DATA – A CASE STUDY: MARITSA RIVER, EDIRNE

A. F. Sunar1,* , N. Yagmur1 , A. Dervisoglu1

1 Istanbul Technical University, Civil Engineering Faculty, Geomatics Engineering Department, 34000, Maslak Istanbul, Turkey - (fsunar, yagmurn, adervisoglu)@itu.edu.tr

KEY WORDS: Maritsa-Evros Basin, Flood, Remote Sensing, Sentinel 1/ 2 Data, GEE

ABSTRACT: A flood, one of the most devastating natural disasters in the world, occurs when water inundates land that's normally dry. Although floods can develop in many ways, river floods (i.e. overflow by rivers or river banks) are the most common. Turkey is one of the flood-affected countries with its 20 main basins in 8 regions. One of the most aggrieved basins in Turkey is the Maritsa river basin in in Eastern Balkans, which also contains the natural border regions with Greece and Bulgaria. 65% of the Maritsa River basin, which originates from the Rila Mountains and joins the Arda and Tundzha rivers, is located in Bulgaria. When the melting snow flow or precipitation in the basin increases, the Maritsa River overflows from the slopes to the Edirne Plain and from time to time exceeds the capacity of the bed, causing floods. On the other hand, since the water level in the dams and reservoirs was kept at the highest level for production purposes, the flood repeat interval increased in the region, since 2000s. Today, it is possible to monitor and evaluate the damages of flood by obtaining very reliable information with space technology. Especially, microwave SAR images that can penetrate clouds, are of great importance in flood mapping because they provide immediate information on the extent of inundation and support the evaluation of property and environmental damages. In this study, rapid flood risk assessment in the region was performed using Landsat 8 and Sentinel 2 Normalized Difference Water Index (NDWI) time series images, and calibrated Sentinel 1 SAR images produced on Google Earth Engine (GEE) platform for 2015-2018 period. GEE is a cloud-based platform that facilitates access to high-performance computing resources to handle very large geographic data sets. The results were compared and verified using meteorological data, riverbed flow data, and digital media news. The results showed that the most affected areas were consistent with the highest measured flow rates and the magnitude of flood damages caused by two main causes in the basin (i.e. opening of shutters in Bulgarian dams or local excessive rainfall) was very different (approximately 8 times larger) from each other.

1. INTRODUCTION operations of the Bulgarian dams without causing economic loss (Sezen, 2007). Floods, which have serious environmental and social impacts worldwide, have been mostly caused by climate-related factors such as severe spring snowmelt and heavy rainfall in the last decade (CRED, 2015).

Turkey is one of the countries affected by the floods with its 20 main basins in eight regions. One of the most aggrieved basins in Turkey is the Maritsa (Meriç in Turkish) river basin, located in Eastern Balkans containing the natural boundary regions with Greece and Bulgaria (INWEB, 2019). River floods in this basin where three rivers meet (biggest tributaries, Tundzha and Arda Rivers, joining in Maritsa in Edirne) are among the most common examples, due to excessive rainfall or snowmelt as well as uncontrolled water released from existing dams in Bulgaria and insufficient riverbed cross-sections. A significant number of reservoirs and cascades (about 2.2 billion m3 on the Maritsa and Tundzha Rivers and about 1 billion m3 on the Arda River in Bulgaria) have been built for irrigation and hydro- Figure 1. Trans-boundary Maritsa river basin and reservoirs electricity production (Figure 1). After 1994, with the (Yildiz, 2011). privatization of dams and hydroelectric power generation, the To prevent natural hazards such as flood, geo-information water level in the reservoirs was kept at the highest level for technologies (i.e. remote sensing and GIS) are used as an production purposes and the flood recurrence interval increased effective tool for risk assessment and hazard management, after this date (Yildiz, 2011). However, uncontrolled water providing a synoptic view and faster analysis than conventional released from existing dams directly overflows on the Maritsa surveying methods. Remote sensing systems with both passive River, causing serious transboundary floods in Edirne. Research optical sensors and active microwave sensors are often used for studies have shown that, especially after 2005, floods, occurred flood mapping and offer different levels of capacity and very often, can be prevented with proper implementations and accuracy (Shen et. al., 2019) (Figure 2). However, since the

* Corresponding author

This contribution has been peer-reviewed. 497 https://doi.org/10.5194/isprs-archives-XLII-3-W8-497-2019 | © Authors 2019. CC BY 4.0 License.

validity of optical observations is limited by clouds, which are study, the period of 2015-2018 was taken into consideration common during flood events, microwave SAR systems are according to the operational dates of Sentinel 2 satellites. preferred providing more efficient analysis due to their penetration capabilities through the atmosphere (Shen et. al., 2019).

Figure 2. Example of satellite dataset used for the latest flood of March 25, 2018. This study contributes to the application of cloud computing Google Earth Engine (GEE) and its potential for flood monitoring and mapping. Recently, GEE is being used for environmental data analysis that can determine important parameters from different satellite time series images. In this study, GEE was used to assess potential flood damage by visualizing and plotting the multi-temporal Landsat/Sentinel 2 NDWI and calibrated Sentinel 1 dB values in time series graphs. The findings were used to assess whether the increase in Figure 3. Map of the study area. flood frequency in the region was due to climate change or improper management of Bulgarian dams, together with Landsat 8 Sentinel 2 Sentinel 1 Satellite ancillary data. (optical) (optical) (SAR)

C band 2. STUDY AREA & DATA USED Spectral 9 Bands 12 Bands (HH,HV,VH, (µm) (0.433-2.30) (0.443-2.20) The Maritsa river basin, is one of the major river systems VV) located in the eastern Balkans, with a total length of 550 km and B1-5,7.9: 30 m B2,3,4,8:10 m a total catchment area of 39,000 km2 (Figure 1) (INWEB, B5,6,7,8a, 2019). The city of Edirne, located in the North West region of Spatial (m) B6: 60 m 5 x 20 m 11,12 : 20 m Turkey, lies at the junction of the Tundzha, Arda and Maritsa rivers, near the borders of Greece and Bulgaria. The area of B8: 15 m B1,9,10: 60 m 44544.78 hectares covering Edirne Center was selected as the Radiometric 6 bit 12 bit 10 bit study area (Figure 3). Since 1571, Edirne city center has Temporal 16 days 10 days 12 days witnessed the most recorded floods resulting significant Resolutions negative environmental and economic impacts with a wide Table 1. The characteristics of the satellite data used variety of consequences from local to national scales (Turoglu (Url-1, Url-2). & Uludag, 2010). Meteorological data and flow measurements were also used as anchillary data in the analysis. Precipitation values and flow Landsat 8 OLI MSI and Sentinel 2 MSI, which are free optical data were obtained from Edirne Meteorological Station and 5 data since 2013 and 2015, were used as a satellite data set different flow observation stations, respectively. Research (Table 1). In addition to optical data, Sentinel 1 SAR data, shows that the discharge in Edirne city center should not exceed which has been free microwave satellite data since 2014, was 1000 m3/s, otherwise it causes flooding. When the maximum used for flood mapping due to its capability of capturing the flow values taken from the measurement station in Edirne center images day/night, irrespective of the weather conditions. In this

This contribution has been peer-reviewed. 498 https://doi.org/10.5194/isprs-archives-XLII-3-W8-497-2019 | © Authors 2019. CC BY 4.0 License.

between 1982 - 2015 are examined; especially after the 2000s, 4. RESULTS it is seen that the critical flow threshold is often exceeded (Url- 3) (Figure 4). 4.1. Optical data analysis

Optical Landsat 8 and Sentinel 2 images were selected from the GEE archive for 2015 - 2018 by filtering with a percentage of cloud coverage of less than 10%. In this context, 52 (from 157) and 145 (from 408) images were used for Landsat and Sentinel 2 satellites, respectively. Then, NDWI time series for this area were created (Figure 5).

Figure 4. Maximum flow values of Maritsa River measured by DSI between 1982-2015 (Url-3).

3. METHODOLOGY

To evaluate the flooding occurance and mapping, the following image processing steps were applied.

- Normalized Difference Water Index (NDWI) NDWI is a widely used remote sensing-derived index to monitor changes related to water content (water level or change in water level ― e.g. flooding) in water bodies. NDWI uses Green and Near-infrared bands of the satellite images to delineate and enhance open water features (Eq. 1) (McFeeters, 1996). NDWI = (Green ― NIR)/(Green + NIR) (1) Figure 5. NDWI time series of the optical images used.

- SAR Calibration & Thresholding Changes in water level can be easily seen in Figure 5, but the The purpose of SAR calibration is to provide images in which main floods (e.g. occurred on January 19, 2016 and March 29, the pixel values may be directly related to the radar backscatter 2018) can be observed within the time series, only if optical of the scene. The unitless backscatter coefficient is converted to images were obtained immediately after the floods. In other dB in order to retrieve an evenly distributed value range from words, optical data may not always be so efficient in flood the non-Gaussian distributed values (Eq. 2). analysis because it is not possible to obtain optical images 0 0 during overflow due to heavy cloud mass. σ (dB) = 10 × log10 σ (2) where, σ0(dB); backscattering image in dB, σ0; Sigma nought 4.2. SAR data analysis image. In this study, 552 Sentinel images were used; and the water and In SAR images, water areas are characterized by low non-water areas were rather well separated in VV polarization backscatter values since the water surface is smooth. From a with thresholds of −20 dB. In other words, all flood risk events statistical point of view, the histogram of water and non-water were extracted from the GEE archive for 2015 – 2018, taking backscattering values is generally characterized by a bimodal into account the calibrated Sentinel 1 Level-1 SAR values of distribution. As mentioned by Uddin et. al, 2019, the radar less than -20 dB (Figure 6); and these images were verified by backscattering characteristics of flood water differ from non- comparing with the data in Turkish Disaster and Emergency water bodies in different SAR polarizations and showed that Management Authority (AFAD) website, and digital media optimum backscatter ranges for inundated areas are between news. As shown in Table 2, some images matched known flood -24 and -17 dB in VH polarization and -22 and -13 dB in VV events, some did not. In addition, flood-affected areas were polarization. Therefore, simple histogram-based thresholding calculated for each event identified in the selected area. was used to extract water areas in each SAR image of different polarization. Table 2 shows that the largest flood-affected area occurred after the flood of 2 February 2015, which was confirmed by AFAD - Google Earth Engine (GEE) and digital media. However, in some floods, digital media and Given the high amount of image download and processing AFAD website do not contain any information Likewise, some required, it is clear that desktop-based systems are not suitable floods take place on digital media, but are not available on the for providing fast rendering services critical to a flood response. AFAD website. It may be concluded from the table that only Today, the Google Earth Engine platform hosts huge dataset to large flood events (approx. average 1500 hectares) causing great analyse and visualize geospatial dataset for different aims. damage, have been reported. Therefore, in this study, a GEE-based approach was used to assess flooding occurances between 2015-2018 using NDWI and calibrated SAR db values.

This contribution has been peer-reviewed. 499 https://doi.org/10.5194/isprs-archives-XLII-3-W8-497-2019 | © Authors 2019. CC BY 4.0 License.

described as “the disaster of the century” due to heavy rainfall in Bulgaria and Greece, led to huge increase in the flow of the Tundzha and Maritsa rivers (Url-5) In other words, as is known, in the event of excessive rainfall or snow melting, the sudden rising flow cannot be carried by the river beds and therefore floods occur, so this flood (i.e. dated on February 2, 2015) is due to the opening of shutters in Bulgaria dams. According to DSI, the water flow reached 2149 m3/sec on the Maritsa River and 346 m3/sec on the Tundzha River that day.

Flow m³/sn Edirne Figure 6. The calibrated Sentinel 1 Level-1 SAR time series Dates Tundzha Maritsa Arda M.S. covering the period 2015 – 2018. Preci. Elh. Sua. Deg. Svil. Kiri. Ivoy. mm/day SAR Image Flood Dates Flood Feb 1, 15 138 252 1052 894 1033 26.3 Dates (Research on Flood Dates Affected Feb 2, 15 371 1234 2101 1570 3.2 (Below -20 dB) Internet ) (from AFAD) Areas (ha) Feb 3, 15 346 1115 2149 860 0.0 Feb 2, 2015Feb 2, 2015 Feb 5, 15 178 251 873 1443 235 0.0 Feb 5, 2015 2323.02 Mar 1,15 127 137 347 582 102 0.0 Mar 2,15 125 138 353 598 111 0.0 March 1, 2015No informationNo information 657.82 Mar 3,15 118 138 342 630 - 0.2 March 8, 2015No information Mar 7,15 122 151 879 880 267 28.4 March 13, 2015 2128.63 Mar 8, 15 141 249 941 1197 275 12.0 March 19 1410.12 Mar 11,15 165 242 1046 1338 381 0.0 Mar 12,15 161 224 1058 1345 396 0.0 March 25,2015March 25, 2015March 25, 2015 1402.19 Mar13,15 156 210 923 1345 339 0.2 March 30,2015 1698.05 Mar 24,15 120 144 549 1147 294 0.0 March 31,2015 1725.91 Mar 25, 15 118 135 526 972 271 0.0 April 11,2015No informationNo information 1327.26 Mar 26,15 115 133 532 953 268 0.0 Mar 27,15 115 131 495 953 270 8.6 April 12,2015No informationNo information 1051.94 Mar 30,15 152 222 817 1338 639 0.1 January 18, 2016 January 18, 2016 Mar 31,15 156 188 750 1356 468 0.2 January 19, 2016 1893.48 Apr 11-12,15 ------0.0 March 8, 2018March 8, 2018No information 745.46 Jan 17, 16 217 733 1000 645 62.4 Jan 18, 16 219 729 1364 683 7.2 March 9, 2018 776.85 Jan 19, 16 110 117 377 1307 637 0.0 March 21, 2018No informationNo information 685.35 Mar 1,18 26 46 52 149 459 43 0.0 March 25,2018March 25,2018 Mar 7,18 85 175 185 406 766 0 1.2 March 26, 2018 936.98 Mar 8,18 85 143 196 363 727 232 0 Mar 9,18 80 132 188 374 739 0 5.2 March 27, 2018 1191.19 Mar 18,18 33 60 78 218 475 241 0.0 April 2, 2018No informationNo information 1314.97 Mar 24,18 101 182 202 421 801 267 6.8 Table 2. Comparison of floods obtained from SAR images with Mar 25,18 103 185 222 413 886 264 3.0 digital media and AFAD data. Mar 26,18 104 180 236 406 875 264 0.2 Table 3 shows the flow values of all overflow stages (i.e. Mar 27,18 111 226 284 614 1109 377 13.2 before, during and after the flood) measured at the flow Mar 28,18 113 215 308 478 1368 440 17.8 monitoring stations and the precipitation data obtained from the Mar 29,18 115 201 264 482 1353 421 9.4 Edirne Meteorological Station. In the table, the dates framed in Mar 30,18 113 194 240 442 1258 388 0.0 blue show the dates of the floods. The Directorate of State Mar 31,18 104 175 219 359 1121 293 0.0 Hydraulic Works (DSI) performs flow measurement many times Apr 1, 18 95 155 204 324 920 249 0.0 a day and they are categorized according to the alarm levels Apr 2, 18 83 135 187 282 801 0 0.0 that displayed in different colors. The alarm levels for each station are determined according to the station's various Nov 20,18 10 14 14 123 145 0 0.0 parameters and are announced when critical values are reached Nov 28, 18 12 16 17 117 152 182 128.5 (Url-4). It is seen from table that no alarm was given by DSI for Nov 29,18 13 15 28 115 289 183 1.8 flood events in March 2015 (listed in Table 2), although large Nov 30,18 13 16 16 115 344 259 0 areas were affected. Alarm Level 1 Alarm Level 2 Alarm Level 3

Flow rates on February 2-3, 2015 and January 17-18, 2016 are the highest flow rates measured by the 11th Regional Table 3. Flows and precipitations measurements. Directorate of Water Affairs. The flood on February 2, 2015,

This contribution has been peer-reviewed. 500 https://doi.org/10.5194/isprs-archives-XLII-3-W8-497-2019 | © Authors 2019. CC BY 4.0 License.

Table 3 shows that the most affected areas due to flood are The Turkish Directorate of State Hydraulic Works performs consistent with the highest measured flow rates. Compared to daily flow measurements and classifies them according to the other flood events, it is seen that the areal extents affected by alarm levels determined according to the various parameters of the flood were also proportional to the river flow rates. the station. As shown in this study, the regions most affected by the flood are consistent with the highest measured flow rates, When the flood recorded on November 28, 2018 is analyzed, it and it appears that the flood affected areas are proportional to is seen that the flow values are quite low. The main reason for river flow rates compared to other flood events. this flood can be explained by the excessive rainfall (128.5 mm/day), which is the highest rainfall since 1930 - 2018. Given This study aims to demonstrate a rapid risk assessment for that the average monthly rainfall is 69.7 mm for the 88-year potential damage assessment by analysing optical and active period, it is seen that the daily precipitation amount at the time satellite image data automatically on the GEE platform. As of the flood was 2 times that of the average value (Url-6). On shown in this study, there are two main causes of floods in the this date, the areal extent of inundated areas was found as 279 basin (i.e. the opening of shutters in Bulgaria dams or local ha. excessive rainfall), and the magnitude of the damages caused by them is different. It is seen that the damage caused by the To compare the inundated areas caused by 2 different main opening of shutters in Bulgaria dams was about 8 times greater reasons in the study area (i.e. the opening of shutters in Bulgaria than the other. dams or local excessive rainfall), 2 images dated 5 February 5, 2015 and November 28, 2018 were taken into consideration Based on the findings of this preliminary study, it can be (Figure 7). concluded that geo-information technologies, including remote sensing, provide prompt information for effective decisions for flood disaster management of cross-border areas such as the Maritsa river basin. Cloud-based computation environments, such as the GEE platform, proved to be particularly valuable for a rapid flood-related emergency response and for assessment of flood damages. For future work, we will consider to use more data available in other cloud computing resources, such as those offered by NASA Earth Exchange (NEX), or Amazon Web Service (AWS).

ACKNOWLEDGEMENTS

We express our gratitude to Sadettin Malkaralı, Edirne Branch Manager of 11th Regional Directorate of State Hydraulic Works, for flow data supply and his guidance.

(a) (b)

Figure 7. Comparison of inundated areas caused by 2 different REFERENCES floods. (a) The opening of shutters in Bulgaria dams (b) Local excessive rainfall. CRED, 2015. The Human Cost of Weather-Related Disasters 1995-2015; Technical Report; Centre for Research on the When two images were examined, it was determined that the Epidemiology of Disasters; UN Office for Disaster Risk flood-affected area as a result of the opening of shutters in Reduction (UNISRD): Geneva, Switzerland. Bulgaria dams was about approximately 8 times larger (2323 ha) than the other event (279 ha). This proves the need for a INWEB (International Network of Water Environment Centres comprehensive agreement with Bulgaria to properly regulate the for the Balkans), Internationally Shared Surface Water Bodies reservoir volumes of these dams. in Balkan Region http://www.inweb.gr/workshops2/sub_basins/13_14_15_Evros_ Ardas_Ergene.html, Accessed August 1, 2019 5. CONCLUSION McFeeters, S. K., 1996. The Use of the Normalized Difference Floods are among Earth's most common–and most destructive– Water Index (NDWI) in the Delineation of Open Water natural hazards. It is known that the Maritsa basin, which is a Features. International Journal of Remote Sensing, vol. 17, no. cross-border and flood-producing river, has a long flood history 7, pp. 1425–32. and the number of floods in this basin have increased dramatically in the last 15 years. Although it is not clear Sezen, N., 2007. River Basin Flood Management: Meriç River whether this increase in flood frequency is due to climate Flood and Turkish-Bulgarian Cooperations, International change (i.e. meteorological causes) or the mismanagement of Congress on River Basin Management. State Hydraulic Works Bulgarian dams, however, it is clear that the drainage system DSI, Turkey. and the carrying capacity of the rivers are not sufficient to prevent floods in Edirne. In this context, as stated by experts, Shen, X., Wang, D., Mao, K., Anagnostou, E., Hong, Y., 2019. the Bulgarian dams on these rivers need to be operated because Inundation Extent Mapping by Synthetic Aperture Radar: A 3 the total flow in Edirne should not exceed 1000 m /s. Review. Remote Sensing, 11(7), 879.

This contribution has been peer-reviewed. 501 https://doi.org/10.5194/isprs-archives-XLII-3-W8-497-2019 | © Authors 2019. CC BY 4.0 License.

Turoglu, H., Uludag, M., 2010. Floods and flashfloods in Edirne (Turkey). International Multidisciplinary Scientific GeoConference: SGEM: Surveying Geology & mining Ecology Management, 2, 9.

Uddin, K.; Matin, M.A.; Meyer, F.J., 2019. Operational Flood Mapping Using Multi-Temporal Sentinel-1 SAR Images: A Case Study from Bangladesh. Remote Sensing, 11, 1581.

Yildiz, D., 2011. Obligation for international cooperation in the water management of the Meriç River basin. Middle East Strategic Research Center, ORSAM Water Research Report, (44).

Url-1: https://landsat.gsfc.nasa.gov/landsat-8/, Accessed July 19, 2019.

Url-2: https://sentinel.esa.int/web/sentinel/home, Accessed July 19, 2019

Url-3: http://www.hidropolitikakademi.org/transboundary- floods-over-meric-maritzaevros.html/3, Accessed July 18, 2019

Url-4: edirnenehir.dsi.gov.tr, Accessed July 19, 2019

Url-5: http://www.hurriyet.com.tr/gundem/edirnede-yuz-yilin- felaketi-28104747, Accessed July 19, 2019

Url-6: https://www.mgm.gov.tr/veridegerlendirme/il-ve-ilceler- istatistik.aspx?m=EDIRNE, Accessed July 16, 2019

This contribution has been peer-reviewed. 502 https://doi.org/10.5194/isprs-archives-XLII-3-W8-497-2019 | © Authors 2019. CC BY 4.0 License.